基于LBP算子和类覆盖捕获图的人脸检测算法

Face Detection Based on Local Binary Pattern Feature and CCCD Classifier

针对传统人脸检测中的过分类问题,提出一种结合LBP算子与类覆盖捕获图的人脸检测算法.该算法首先用ε-LBP算子提取人脸图像纹理特征,并把对应不同ε值提取的LBP特征数据加权融合起来,形成人脸图像特征向量,然后采用类覆盖捕获图构造分类器,最终对人脸图像实现有效检测.与传统方法相比,基于随机图理论的类覆盖捕获图能够克服过分类缺陷,比其他近邻图分类器更具优势,性能也比较稳定.实验结果表明,该算法可以有效检测人脸图像,尤其对存在模糊和光照异常的人脸图像具有较高的精确度和鲁棒性.

Aiming at the over-classification drawback, this paper presents a use of improved- LBP operator and joins Class Cover Catch Digraphs （CCCD） classifier for face detection. This operator with-LBP texture features extracted from face images, and to correspond to different values of the extracted face feature data weighted fusion, it will be weighted data fusion to describe the characteristics of the face. Combing with the CCCD classifier, the face images can be detected at last. Compared to the traditional methods, the CCCD classifier, based on random graph theory, can overcome the over-classification drawback efficiently. It＇s more stable than the other nearest neighbor graph classifier. The experimental results show the proposed algorithm can achieve better robustness, higher accuracy than the traditional methods, especially for that for the real-world blurry and illumination changing images. It is an effective automatic face detection method.